impact of champ radio occultation observations on global ...250 k (hajj et al. 2003). the new champ...

$: Impact of CHAMP Radio Occultation Observations on Global ...250 K (Hajj et al. 2003). The new CHAMP re-ceiver substantially increases the fraction of GPS occultations that reach the$
Journal of the Meteorological Society of Japan, Vol. 82, No. 1B, pp. 533--549, 2004 533

Impact of CHAMP Radio Occultation Observations on Global Analysis

and Forecasts in the Absence of AMSU Radiance Data

X. ZOU, H. LIU

Florida State University, Tallahassee, Florida, USA

R.A. ANTHES

University Corporation for Atmospheric Research, Boulder, Colorado, USA

H. SHAO

Florida State University, Tallahassee, Florida, USA

J.C. CHANG

Chinese Culture University, Taipei, Taiwan

and

Y.-J. ZHU

National Centers for Environmental Prediction, Camp Springs, Maryland, USA

(Manuscript received 13 June 2003, in revised form 11 November 2003)

Abstract

Challenging Minisatellite Payload (CHAMP) radio occultation (RO) observations during a two-weekperiod are assimilated into global analyses using the National Center for Environmental Prediction(NCEP) three-dimensional variational (3D-Var) system with a recently improved observation operatorfor assimilating GPS bending angle data. The NCEP 3D-Var system used in this research is suboptimalsince Advanced Microwave Unit (AMSU) radiances are not included in our experiments. Analyses withand without CHAMP observations are compared with each other and with collocated conventional ra-diosonde and dropsonde data, which are excluded from both experiments. Zonal mean temperature, hu-midity and surface pressure differences between the GPS analyses and NO-GPS analyses are examined.The GPS analyses in the Southern Hemisphere show higher temperatures than the NO-GPS analyses,particularly in the mid- and high latitudes. The GPS analyses show drier air in the lower troposphereand more moist air in the middle troposphere compared to the NO-GPS analyses. The surface pressure isslightly increased (maximum 0.8 hPa) in the Southern Hemisphere and decreased (maximum 0.25 hPa)

Corresponding author: Xiaolei Zou, Departmentof Meteorology, Florida State University, 404,Love Bldg, Tallahassee, Florida, 32306-4520,USA.E-mail: [email protected]( 2004, Meteorological Society of Japan

in the Northern Hemisphere due to the inclusion of GPS observations. Compared with the collocated in-dependent soundings, the large cold bias (as large as 2.5 K) in the NCEP Southern Hemisphere analysesproduced without CHAMP observations is significantly reduced. On average, a 20% mean error reduc-tion in the temperature analysis is obtained in the Southern Hemisphere when CHAMP data are in-cluded. Degradations in the surface pressure analysis found from previous the GPS/Meteorology dataassimilation studies are greatly reduced. The differences between the surface pressure analysis errorswith and without CHAMP data are less than 0.8G1.5 hPa. Comparisons of numerical forecasts ini-tialized with analyses produced with and without CHAMP occultations display a small improvement inthe forecasts in the tropics and the Southern Hemisphere associated with the use of the CHAMP ob-servations.

1. Introduction

Following the pioneering satellite missionthat demonstrated the Global Positioning Sys-tem (GPS) radio occultation (RO) technique toproduce global soundings of the Earth’s atmo-sphere—the GPS/Meteorology (GPS/MET) ex-periment (Ware et al. 1997), the German/USCHAMP (Challenging Minisatellite Payload)mission was launched on July 15, 2000, and ROmeasurements from CHAMP began on Febru-ary 11, 2001 (Wickert et al. 2001). CHAMPcarries the latest generation JPL (Jet Pro-pulsion Laboratory) GPS receiver, BlackJack,which provides improved signal quality and al-lows for application of advanced signal trackingtechniques (Yunck et al. 2000). The penetrationof occultations into the lowest 1/2 km of the at-mosphere over the entire globe has increasedfrom about 30% for the GPS/MET experimentto almost 50% for CHAMP (Hajj et al. 2003).Having high vertical resolution (0.1–1 km)and high accuracy (< 1 K for temperature,0.5 g kg�1 for specific humidity) under virtuallyall weather conditions (Hajj et al. 2003), GPSoccultations are unique and complement othersatellite observations, which often have highhorizontal resolution, but lower vertical resolu-tion. Assimilation of GPS RO measurements inglobal analyses has great potential to contrib-ute to climate monitoring, weather forecastingand atmospheric research.

Considerable effort has been made to developtechniques for assimilating RO data into nu-merical weather prediction (NWP) models.Various strategies for assimilating data werediscussed by Eyre (1994) and Kuo et al. (2000).Kursinski et al. (2000) combined occultationrefractivity profiles from GPS/MET data withECMWF (European Centre for Medium-RangeWeather Forecasts) global analyses in a 1D-Var

framework. Healy and Eyre (2000) tested thesame strategy in a simulated approach. Andmore recently, Poli et al. (2002) explored thisoption with the finite volume DAO’s DAS back-ground. Palmer et al. (2000, 2001) proposed toassimilate GPS bending angle data using a for-ward Abel transform. Von Engeln et al. (2003)basically took the same route to probe the re-trieval sensitivity with a 1D-Var tool. Zou et al.(1999, 2000), Liu et al. (2001), and Shao andZou (2003) used a ray-tracing procedure for theassimilation of GPS/MET bending angles. Ithas been found that the moisture analysis,after the assimilation of GPS/MET refractivityprofiles, is in general drier than the originalanalysis (Kursinski et al. 2000; Palmer et al.2001). Interestingly, the comparisons with ra-diosondes can either be wetter (Palmer et al.2001) or drier (Poli et al. 2002), with differentbackgrounds and/or approaches. Large profile-to-profile temperature discrepancies were alsofound in the tropical lower stratosphere due toreal waves (probably gravity waves, e.g., Tsudaet al. 2000) that were not represented in theECMWF analysis (Kursinski et al. 2000). Re-sults from assimilating GPS/MET bending an-gles using a ray-tracing procedure by Zou et al.(2000), Liu et al. (2001), and Shao and Zou(2003) are summarized in Section 3.1.

Following the successful demonstration ofthe assimilation of GPS/MET data, severalchallenges remain: (i) obtaining a statisticalevaluation on the impact of RO measurementsto global analyses and forecasts, (ii) assess-ing the added value of RO observations toglobal analyses and forecasts in the presence ofother observations, (iii) improving the mois-ture analysis in the lower troposphere withRO observations, (iv) eliminating the degrada-tion in the surface pressure analysis from ROassimilation, and (v) developing a fast and

Journal of the Meteorological Society of Japan534 Vol. 82, No. 1B

operationally-feasible method for the assimila-tion of RO occultations.

This research extends previous RO dataassimilation studies to the assimilation ofCHAMP data and addresses issues (i)–(iv)above. Work on the fifth challenge will be pre-sented in a separate paper. Major differencesbetween this study and our previous GPS dataassimilation studies include the use of a higherresolution model (T170L42 instead of T62L28),CHAMP observations, other observations (ex-cluding AMSU data), and a more accurate ray-tracing operator. The purpose of this effort is toassess the added value of RO observations toglobal analyses and one- to five-day forecasts.

In Section 2 we provide a brief description ofCHAMP occultations. Section 3 provides anoverview of the three-dimensional variational(3D-Var) data assimilation algorithm and theexperimental design. In Section 4 the zonal andmeridional distributions of GPS-derived tem-perature and moisture increments are pre-sented, and analyses produced with and with-out CHAMP occultations are compared withcollocated conventional observations, which areexcluded from both data assimilation experi-ments. Forecasts from analyses produced withand without CHAMP occultations are eval-uated in Section 4, and the summary and con-clusions are given in Section 5.

2. CHAMP GPS radio occultationmeasurements

GPS RO measurements are based on thetime delay of the transmitted GPS radio wavesfrom a GPS satellite occulted behind the Earthas viewed from a low-Earth-orbiting satellite(Fjeldbo et al. 1971). The magnitude of the de-lay depends on the density of the Earth’s atmo-sphere. Precise time delay measurements areconverted first into an atmospheric Dopplershift measurement and then to a bending anglemeasurement based on satellite geometry (Mel-bourne et al. 1992). The bending angles can beused in a data assimilation system to improveglobal analyses and forecasts (Zou et al. 1999).

Improvement to global analyses and fore-casts is determined by not only how RO obser-vations are assimilated, but also by RO dataavailability and quality. Using a new genera-tion of ‘‘semi-codeless’’ GPS receivers, accuratetracking of the GPS signals can be made by the

CHAMP receiver while Anti-Spoofing (AS) isturned on, i.e., when the GPS signal is en-crypted by the Department of Defense. Errorsin individual temperature profiles are found tobe less than 0.5 K in dry regions where T <250 K (Hajj et al. 2003). The new CHAMP re-ceiver substantially increases the fraction ofGPS occultations that reach the lower tropo-sphere. The CHAMP mission also providesmore daily occultations than the GPS/MET ex-periment and hence greater spatial coverage.

We use CHAMP observations from July 16th

to July 31st, 2002 in this study. Details aboutthe CHAMP data and their processing can befound from the UCAR (University Corporationfor Atmospheric Research) COSMIC (Constel-lation Observing System for Meteorology, Iono-sphere and Climate) Data Analysis and Ar-chival Center (CDAAC) website (http://www.cosmic.ucar.edu:8080/cdaac/index.html). A totalof 3030 occultations were processed and madeavailable during this two-week period. The dis-tribution of the processed data over this periodis shown in Fig. 1. On average, about 200 oc-cultations per day pass the internal qualitycontrol (QC) imposed by the UCAR CDAACRO retrieval procedure. This QC procedure in-cludes the truncation of the L1 signal, thetruncation of the L2 signal while keeping theL1, and the noise reduction of bending anglesat high altitudes (primarily above 30 km) usingthe CIRAþQ climatological data (Kirchengastet al. 1999). A more detailed description aboutthe internal QC can be found in Kuo et al.(2003). Of those 3030 CHAMP soundings, 1427CHAMP soundings (47% of the total) pene-trated to a level below 850 hPa. No further QChas been done for the acquired bending angleprofiles in our subsequent data assimilationexperiments.

Information provided by CDAAC CHAMPlevel-2 data that are used for bending angle as-similation is listed in Table 1. The main differ-ence between CHAMP level-2 data and GPS/MET level-2 data is that the normal direction ofthe occultation plane, directly provided in theGPS/MET level-2 data set, is not provided inthe CHAMP level-2 data set. Instead, the azi-muth angle (denoted by the parameter ‘‘az’’ inTable 1) is provided. However, the ray-tracingobservation operator for direct assimilation ofbending angle data requires the normal direc-

X. ZOU, H. LIU, R.A. ANTHES, H. SHAO, J.C. CHANG and Y.-J. ZHU 535March 2004

tion of the ray at the perigee point, which isused to derive the tangent direction for the rayintegration (see Section 2 in Zou et al. 1999).Calculation of the normal direction, un

occ, basedon CHAMP level-2 data, is therefore required.Details of this derivation are presented in Ap-pendix A.

3. Assimilation of GPS bending anglesand experimental design

3.1 A brief review of GPS RO assimilationThe authors’ efforts for assimilating GPS RO

data started in 1993, before the GPS/METlaunch, which occurred on April 3, 1995 (Wareet al. 1996). Zou et al. (1995) and Kuo et al.(1997) conducted two sets of observing systemsimulation experiments to assess the potentialimpact of GPS refractivity data on temperatureand moisture field analysis and short-rangenumerical weather prediction at meso- and syn-

optic scales, respectively. They found beneficialimpacts from assimilating refractivities for theprediction of the blizzard event of March 8,1992 and an extratropical cyclone known as theERICA (Experiment on Rapidly IntensifyingCyclones over the Atlantic) IOP (Intensive Ob-serving Period) 4 storm which took place overthe Northwestern Atlantic Ocean on January4–5, 1989. Zou et al. (1999) began the develop-ment of a ray-tracing method for incorporatingGPS bending angle measurements directly intoa numerical weather analysis system in orderto handle the complicated nonlinear depen-dence of the RO observations on the atmo-spheric refractivity along the ray path. By di-rectly assimilating either bending angles orrefractivities, the ambiguity between the effectsof water vapor and temperature can be avoided.They then linked the ray-tracing model and itstangent linear and adjoint operators into theSpectral Statistical Interpolation (SSI) analysissystem of the National Centers for Environ-mental Prediction (NCEP). The new bendingangle assimilation system was tested usingover thirty actual GPS/MET observations (Zouet al. 2000). Liu et al. (2001) assessed the im-pact of assimilating the GPS bending angle on

(a)

-90 -60 -30 0 30 60 90Latitude

0

2

4

6

8

10

Low

est p

erig

ee p

oint

hei

ght (

km)

(b)

Fig. 1. (a) The locations of the 3030CHAMP occultation profiles from July16th to July 31st, 2002. (b) The lowestaltitudes of the occultation profiles.

Table 1. GPS level-2 information usedfor bending angle assimilation.

Data Description

start_time Starting time for theoccultation

stop_time Ending time for the occultation

year The year during which thisoccultation ended

day The day of year during whichthis occultation ended

MSL_alt Mean Sea Level altitude

impact_parm Impact parameter

bend_ang Bending angle at perigee

lat Longitude of perigee point

lon Longitude of perigee point

az Angle from the occultation planin the direction of the GPSsatellite with North


analyses and forecasts by incorporating 837GPS/MET occultations for the period June 20–30, 1995. The impact on the global analysis wasexamined by comparing the analysis producedwith the bending angles with 56 collocated ra-diosonde profiles. Analyses of the tempera-ture and specific humidity above 850 hPa werecloser to radiosonde measurements than theoriginal guess fields. They also found that in-cluding the bending angles resulted in a smallimprovement in the short-range (6-h) andmedium-range (one to five days) temperatureforecast, especially in the Southern Hemi-sphere.

After this preliminary work on GPS data as-similation, further efforts were made to im-prove the accuracy and efficiency of GPS bend-ing angle assimilation. Zou et al. (2002) carriedout a statistical estimate of the errors in thecalculation of bending angles caused by a 2Dapproximation of ray tracing and the assump-tion of spherical symmetry of the atmosphere.Shao and Zou (2003) studied the sensitivity ofassimilating GPS/MET bending angles to thespecification of observational weightings. Theyfound that the temperature and the surfacepressure are more sensitive than moisture tothe specification of weightings. Liu and Zou(2003) improved the accuracy and efficiency of a2D forward GPS ray-tracing model. The pres-ent work is based on the improved bending an-gle assimilation system.

3.2 Experimental designWe use the SSI analysis system of the NCEP

for this study. Zou et al. (2000) describe themathematical formulation of the 3D-Var prob-lem. An incremental approach is used in whichthe inner loop solves a linear equation for theincrement added at each outer loop using astandard linear conjugate-gradient algorithm(Parrish and Derber 1992). Data assimilationexperiments are carried out at a horizontalresolution of T170 (approximately 100 km atthe equator). In the vertical, there are a total of42 s-layers, with the model top located at about2 hPa (40 km). Data assimilation is conductedat 6-h intervals. At each analysis time, twoouter loops and 200 inner loops (100 for eachouter loop) are carried out.

As was explained in Zou et al. (1999), GPSbending angles, as a function of the impact pa-

rameter aðaÞ, are incorporated directly into theNCEP SSI system. Starting from the samebackground field at 00 UTC July 16, 2002, twodata assimilation cycles are conducted. Thefirst data assimilation cycle, NO-GPS, includesvirtually all observations that were assimilatedoperationally at NCEP during that time pe-riod, except for the NOAA-16 AMSU radiancedata1. The observations used are summarizedat Table 2. The 6-h forecast initialized by theprevious analysis is used as a guess field for thenext analysis. The second data assimilation ex-periment, GPS, is the same as NO-GPS except

Table 2. The observations included inthe NO-GPS experiment (after pass-ing the SSI internal quality controland thinning procedures) on 00 UTC,July 16, 2002.

Data TypeTotal number

of observations

NOAA-15 HIRS 1b radiance 33978

NOAA-14 HIRS 1b radiance 27391

NOAA-15 AMSU-A 1bradiance

58629

NOAA-15 AMSU-B 1bradiance

103766

NOAA-14 MSU 1b radiance 5442

NOAA-16 SBUV ozoneprofile

187

GOES-8 5 � 5 cloud clearedradiance

6559

GOES-8 5 � 5 cloud clearedradiance

4724

q 10701

ps 13451

T 53966

u; v 137625

1 These data were inadvertently omitted in ourexperiments. Omission of these data results inlower forecast skill scores than the NCEP opera-tional forecasts and probably results in an over-estimation of the impact of the CHAMP RO ob-servations compared to the impact if the AMSUdata were included in the experiments.


that CHAMP bending angle observations areadded.

A total of 229 out of 3030 CHAMP profilesoccurred within a 3-h time window and a 200-km distance of conventional soundings (radio-sonde and dropsonde profiles). These collocatedconventional soundings, which are excludedfrom the assimilation in both experiments, areintended for an independent verification. Sincethe vertical resolution of the data assimilationsystem is much lower than that of the CHAMPmeasurements, the CHAMP bending angles areassimilated at selected levels, with an intervalof 0.2 km below 3.2 km, 0.4 km between 3.2and 10 km, 1 km between 10 and 20 km, and2 km above 20 km. The average number of rayssimulated for each occultation is 69.

The natural logarithm of bending angle (ln a)is used as the GPS observational quantity inthe cost function J as in Zou et al. (2000). Theweighting coefficients for the GPS data assimi-lation procedure are calculated as the inverseof the mean square difference between the ob-served and the simulated ln a, calculated fromthe background field-using a total of 4306CHAMP occultations in May 2002. The meansquare differences in terms of ln a are shown inFig. 2. The error variances compare favorablywith the total bending angle errors estimatedby Palmer et al. (2000). As tested in Shao andZou (2003), such an estimate of the observationerror variance based on observational incre-ments produces a reasonably good result.

The observation operator for the bending an-gle assimilation consists of a ray-tracing model,whose basic physical concepts and numericalprocedures were described in Zou et al. (1999).Several improvements made to the ray-tracingmodel were presented in Liu and Zou (2003). Inthe ray-tracing model, the Earth is assumedspherical with a single radius ðrlocÞ of local cur-vature for each occultation. At the tangentpoint, rloc is related to the impact parameterthrough the following relationship:

at ¼ nðzMSL þ rlocÞ ð1Þ

where n is the index of refraction and zMSL isthe tangent point mean sea level height. At andabove a height of 40-km height, nA1. There-fore, the radius of the local curvature of theEarth for each occultation is calculated in theforward ray-tracing model as the difference be-tween the impact parameter and the altitude ofthe ray:

rloc ¼ atj40 km � ZMSL ð2Þ

where at and zMSL are provided in the CHAMPlevel-2 data. This ensures the consistency inthe calculation of the local curvature radius ofthe Earth between the ray-tracing model andthe CHAMP data and therefore eliminates theneed for an impact parameter offset that waspresent in our previous studies (Liu et al. 2001;Liu and Zou 2003). The ray integration startsfrom the observed perigee point and is carriedout in two opposite directions, one toward theLEO satellite and the other toward the GPSsatellite. This proves to be a more accuratescheme than the original scheme in which thesimulated ray path starts from a virtual GPSposition (Liu and Zou 2003). The newer tech-nique ensures that the impact parameter andthe tangent direction of a simulated ray at theperigee point, where the most bending of theray occurs, are closer to the actual measure-ments than in the original technique.

4. Numerical results

We have combined the background state vec-tors and all other observations with and with-out bending angle profiles from CHAMP occul-tation observations from July 16th to July 31st,2002. The background state vectors are gen-erated by 6-h forecasts using the NCEP globalspectral model at T170L42 resolution. We first

Fig. 2. Vertical variation of the esti-mated variance of the logarithm ofbending angles.


discuss the results in terms of temperature,specific humidity and surface pressure differ-ences between the 3D-Var solutions with andwithout CHAMP GPS occultations. The 3D-Varresults are then compared with collocated ra-diosonde and dropsonde observations in termsof temperature, specific humidity and surfacepressure. The differences are expressed by themean and the standard deviation on latitudeversus height and/or longitude versus heightcross sections.

4.1 Analysis differencesFigure 3 shows the zonal mean temperature

differences between GPS and NO-GPS analy-ses averaged over the two-week period. Thesmaller temperature differences between the

solution with and without CHAMP occultationsin the Northern Hemisphere in Fig. 3 reflectsthe relatively smaller temperature errors in theNorthern Hemisphere analysis compared to theSouthern Hemisphere analysis in the absenceof CHAMP occultations. The GPS analyses aresignificantly warmer than the NO-GPS analy-ses south of 30�S in the entire troposphere (ex-cept in a cold pocket between 45–85�S and ap-proximately 300–500 hPa). Large temperatureincreases in the GPS solution are found in themiddle and high latitudes. The maximum isover 2.1 K located around 80�S near the sur-face. It should be noted that the variation ofsurface height is very large around this lat-itudinal circle. There is a small low-latitudecooling impact between s ¼ 0:3 and s ¼ 0:15 inthe GPS, indicating a better representation ofthe extremely cold temperatures at the tropo-pause by the high vertical resolution of GPSobservations compared to the other satelliteobservations that are used in the analyses.Above the tropopause in the tropics, a weaksystematic warming is also observed. Exceptfor the lower troposphere near the north pole,the mean temperature changes introduced byCHAMP occultations are less than 0.3 K in theNorthern Hemisphere. The standard deviationsof the temperature changes in the SouthernHemisphere are also larger than in the North-ern Hemisphere. The region of large standarddeviations coincides with a region of largewarming produced by the GPS observations inthe Southern Hemisphere, indicating signifi-cant disagreement between the model and GPSobservations in individual profile structures aswell as their average. These results indicate alikely significant operational impact of RO ob-servations in the Southern Hemisphere.

The average zonal mean specific humiditydifferences between GPS and NO-GPS analysesare shown in Fig. 4. The GPS solution is drierthan the NO-GPS solution throughout the tro-posphere below 700 hPa and north of 30�Swhere the water vapor content is substantial inthis season. The locations of the greatest watervapor changes are different from those of thetemperature changes (compare Figs. 3 and 4).The maximum drying impacts of more than0.2 g kg�1 occur near the north pole, 850 hPa,and in the northern subtropics around 900 hPa.The drying impact generally increases toward

(a)

(b)

Fig. 3. (a) Zonal mean temperature dif-ferences between GPS and NO-GPSanalyses averaged over the two-weekperiod. (b) Zonal mean standard devia-tions of temperature difference betweenGPS and NO-GPS analyses. The con-tour interval is 0.3 K in (a) and 0.2 K in(b). Areas with valuesa�0:1 K in (a)are shaded.


low latitudes. These results are consistent withthose of Kursinski and Hajj (2001), who foundthat the GPS/MET RO observations below6 km are generally drier than the ECMWFanalyses, with the difference generally increas-ing toward warmer temperatures. Above about700 hPa, the GPS analysis is wetter than theNO-GPS analysis, with the largest increase ofspecific humidity (> 0.1 g kg�1) primarily lo-cated in the Southern Hemisphere between0�–60�S and just north of the equator near650 hPa.

Kursinski and Hajj (2001) found that theGPS/MET observations are more moist thanthe NCEP analyses in the lower troposphere oflow latitudes (10 S–30 N), which is not seen inour GPS analyses (Fig. 4). Since the moisteningimpact of the GPS/MET observations in thisregion was not found with the ECMWF analy-ses by Kursinski and Hajj (2001), we believethis difference in the GPS impact on the NCEP

analyses between our study and that of Kur-sinski and Hajj (2001) in this region resultsfrom the reduced dry bias in the NCEP analy-ses since 1995, as well as the limited numberof GPS/MET profiles probing the low-latitude,near-surface environment. In addition, differ-ent approaches used in both studies may alsocontribute to this disagreement. Standard de-viations are largest at the equator, reaching apeak value of more than 0.3 g kg�1, near 5�Sand 700 hPa.

Figure 5 shows the latitudinal dependenceof the zonal mean differences between GPS

(a)

(b)

Fig. 4. Same as Fig. 3 except for spe-cific humidity. The contour interval is0.05 g kg�1 in (a) and 0.1 g kg�1 in (b).Areas with valuesa�0:05 g kg�1 in (a)are shaded.

(a)

(b)

Fig. 5. (a) Latitude dependence of themean surface pressure differences be-tween GPS and NO-GPS analyses aver-aged over the two-week period. (b)Latitude dependence of zonal meanstandard deviations of surface pressuredifference between GPS and NO-GPSanalyses. The solid line is for averageover all longitudes, the dashed line isfor the average from 180�E to 120�W,and the dot-dashed line is from 70� to130�E. Also indicated in (a) is the long-dashed zero line.


and NO-GPS surface pressure analyses. In theNorthern Hemisphere, where there are rela-tively many radiosonde and surface observa-tions, the GPS analysis is close to the NO-GPSanalysis. In the Southern Hemisphere, themean surface pressure from the GPS solution islarger than that of the NO-GPS solution. Thelarger magnitude of surface pressure impactin the Southern Hemisphere compared to theNorthern Hemisphere is a result of a largerbias in the background bias in the SouthernHemisphere. The standard deviations of thesurface pressure differences between the GPSand NO-GPS solutions (Fig. 5b) reach a maxi-mum of approximately 1.0 hPa, which is con-sistent with the surface pressure analysis errorestimate.

In order to examine the effect of the denseradiosonde network in the Northern Hemi-sphere, which imposes additional constraintson the surface pressure analysis, we calculatedthe surface pressure analysis differences be-tween GPS and NO-GPS over a data-void areain the Pacific (180�W–120�E) and a data-denselongitudinal band (70�E–130�E) (Fig. 5). Re-

sults in Fig. 5 indicate that the mean surfacepressure difference between GPS and NO-GPSis very small in both longitudinal bands in theNorthern Hemisphere.

4.2 Analysis verification with collocatedradiosonde and dropsonde observations

Out of 3030 CHAMP occultations, there are229 occultations with at least one collocatedradiosonde and/or dropsonde profile that hasmore than five levels of data and is observedwithin a G3-h time window and a 200-km dis-tance. In order to conduct an independent veri-fication for the bending angle assimilation, weexcluded these collocated radiosonde and drop-sonde observations (a total of 264) from bothexperiments (NO-GPS and GPS). The locationsof these CHAMP occultations are shown in Fig.6a; the lowest perigee point heights of these oc-cultations are shown in Fig. 6b; and the lat-itudinal distributions of collocated radiosondesin time are shown in Fig. 6c. Most of the con-ventional observations selected for verificationare located in the middle latitudes of theNorthern Hemisphere. Many of the CHAMP

(a)

16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31Date of July

90S70S50S30S10S10N30N50N70N90N

00Z06Z12Z18Z

(c)

-90 -60 -30 0 30 60 90Latitude

0

2

4

6

8

10

Low

est p

erig

ee p

oint

hei

ght (

km)

(b)

Fig. 6. (a) The locations of the 229CHAMP occultation profiles collocatedwith conventional data. (b) The lowestaltitudes of the collocated CHAMP oc-cultation profiles. (c) The latitudinaldistribution of the collocated conven-tional data in time.


occultations reach to very low altitudes (atleast 30 reached below 1.0 km). For compar-isons between the analyses and conventionalobservations, the analyses are interpolated tothe locations of observations.

The mean and standard deviations of the dif-ferences between the temperature and specifichumidity analyses obtained by NO-GPS andGPS and the independent soundings are pre-sented in Figs. 7–8. Comparisons are made fortwo areas: the relatively data-rich North Amer-ica (15–90�N, 40–120�W) and the relativelydata-poor Southern Hemisphere (10–90�S, 0–360�E). In the Southern Hemisphere, analyseswithout CHAMP data are systematically colderthan conventional observations below 400 hPa(solid line in Fig. 7a). The cold bias in the GPSanalyses is significantly reduced (a 40% reduc-tion). In the upper troposphere, the analyseswithout GPS observations are slightly warmerthan conventional observations. This warmbias is reduced slightly when GPS observationsare assimilated. The GPS analyses near the top(0–50 hPa) of the atmosphere are colder thanthe NO-GPS analyses and conventional obser-vations, which is probably a result of a GPSobservational bias at these high altitudes. Thedifferences between NO-GPS and GPS temper-ature analyses in North America (Fig. 8) are

much smaller than in the Southern Hemi-sphere (Fig. 7). Both the biases and standarddeviations of the differences are reduced by as-similating GPS observations.

The impact of CHAMP data on the water va-por analyses in the Southern Hemisphere andNorth America varies with altitude (Fig. 7b and8b). The results are difficult to interpret be-cause the radiosondes tend to have a dry biasin the upper troposphere. Furthermore, watervapor can vary greatly over spatial scalessmaller than 200 km, and so radiosondes canhave significant representativeness errors com-pared to large-scale analyses (Kuo et al. 2003).However, the mean differences in all cases aresmall (generally less than 0.25 g kg�1). AddingCHAMP observations in both the SouthernHemisphere and North America increases themean amount of water vapor in the analysescompared to the radiosondes above 800 hPaand dries the analyses below this level. Thestandard deviations of the differences in theanalyses compared to radiosondes are reducedat all levels, except near the surface. The nega-tive N-bias found in RO observations in themoist lower troposphere (Rocken et al. 1997;Kuo et al. 2003) may contribute to the observeddrying of the analyses in the lower troposphere.Future GPS RO observations with the adoption

0.8 1.2 1.6 2 2.4Temperature (K)

0

200

400

600

800

1000

Pre

ssur

e (h

Pa)

-3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1Temperature (K)

0

200

400

600

800

1000

Pre

ssur

e (h

Pa)

(a)0 0.2 0.4 0.6 0.8 1

Specific humidity (g/kg)

300

400

500

600

700

800

900

1000

Pre

ssur

e (h

Pa)

-0.5 -0.25 0 0.25 0.5Specific humidity (g/kg)

300

400

500

600

700

800

900

1000

Pre

ssur

e (h

Pa)

(b)

Fig. 7. (a) The mean (upper panel) and the standard deviation (bottom panel) of temperature differ-ences between the analyses and the collocated radiosonde observations in the Southern Hemi-sphere. The solid line is for NO-GPS analyses and dashed line for GPS analyses. (b) Same as (a)except for specific humidity.


of open-loop tracking and the removal of thenavigation modulation from the GPS signal willsignificantly reduce such a bias (Ao et al. 2003).

4.3 Forecast results with and withoutCHAMP occultations

Five-day forecast experiments initializedwith GPS and NO-GPS analyses at 12 UTCfrom July 16th to July 31st, 2002 are carried outusing the NCEP global spectral model atT170L42 resolution (same resolution as thedata assimilation experiments). Figure 9 showsthe root-mean-square (RMS) error of geopoten-tial height at 500 hPa, specific humidity at850 hPa and surface pressure in the SouthernHemisphere. Figures 10–11 show the anomalycorrelation of geopotential height at 500 hPain the Southern Hemisphere. Figure 12 pres-ents the RMS error of geopotential height at500 hPa and specific humidity at 850 hPa inthe Northern Hemisphere tropics (0–30�N).Results in the Southern Hemisphere are calcu-lated over traditional verification areas (i.e.,20–80�S) following the NCEP’s standard proce-dure. The GPS analyses are used as verifica-tion dataset for the RMS calculations and theNCEP climatology is used for anomaly calcula-tion. Marginal, but discernable improvementsin the forecast skill are seen in both hemi-spheres. A further breakdown in terms ofspatial scales in AC scores reveals that the

0.6 0.8 1 1.2 1.4 1.6 1.8 2Temperature (K)

0

200

400

600

800

1000

Pre

ssur

e (h

Pa)

-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1Temperature (K)

0

200

400

600

800

1000

Pre

ssur

e (h

Pa)

(a)0 0.5 1 1.5 2

Specific humidity (g/kg)

300

400

500

600

700

800

900

1000

Pre

ssur

e (h

Pa)

-0.5 -0.25 0 0.25 0.5Specific humidity (g/kg)

300

400

500

600

700

800

900

1000

Pre

ssur

e (h

Pa)

(b)

Fig. 8. (a) Same as Fig. 7a except for the North America area (15�–70�N, 40�–120�W). (b) Same asFig. 7b except for the North America area.

1 2 3 4 5Time (day)

2

4

6

8

10

RM

S e

rror

(hP

a)

1 2 3 4 50.5

0.7

0.9

1.1

1.3

1.5

RM

S e

rror

(g/

kg)

1 2 3 4 520

40

60

80

100

RM

S e

rror

(m

)

(a)

(b)

(c)

Fig. 9. The RMS errors of (a) 500 hPageopotential height, (b) 850 hPa specifichumidity, and (c) surface pressure fore-casts in the Southern Hemisphere fromone to five days. The solid line is for theNO-GPS forecasts, and the dashed lineis for the GPS forecasts.


improvements mainly come from a better de-scription of the evolution of synoptic scales. Inthe daily performance scores shown in Fig. 11,the advantage of including the GPS observa-tions stands out much more clearly after aboutone week, which we consider to be the forecastanalysis spin-up time period. No substantialimpact is found in the Northern Hemispherenorth of 30�N (not illustrated).

In order to examine the impact of 6-h contin-uous cycling, we have plotted the daily varia-tion of the RMS error of the 500-hPa height forfive forecast days from July 16th to July 31st,2002 in the Southern Hemisphere (Fig. 13). Amore significant impact of CHAMP RO obser-vations is observed during the second week ofthe continuous data assimilation cycling thanwithin the first week of the cycle, especially forthe day-1 and day-2 forecasts. In other words,with the current CHAMP data density, it seemsto take about one week for the analysis to be‘‘spun-up’’ and to show an impact on the South-ern Hemisphere.

The GPS analyses are used for verification offorecasts shown in Figs. 9–13. Figure 14 showsthe rms error of the five-day forecasts ini-tialized with both the NO-GPS and GPS analy-ses verified with the NCEP operational analy-ses. The rms error of the NCEP operationalforecasts is also shown in this figure (dash-

0 1 2 3 4 5Forecast days

0.5

0.6

0.7

0.8

0.9

1

Ano

mal

y co

rrel

atio

n

0 1 2 3 4 50.5

0.6

0.7

0.8

0.9

1

Ano

mal

y co

rrel

atio

n

(a)

(b)

Fig. 10. The anomaly correlations of500 hPa geopotential height forecastsin the Southern Hemisphere from oneto five days: (a) Wave number 4–9, and(b) all waves (waves 1–20). The solidline is for the NO-GPS forecasts, andthe dashed line is for the GPS forecasts.

21JUL 22 23 24 25 26 27 28 29 300.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Ano

mal

y co

rrel

atio

n

EXPctl=0.693EXPgps=0.702

Fig. 11. Daily variation of the anomalycorrelations of 500 hPa geopotentialheight forecasts in the Southern Hemi-sphere at day five (all waves from 1 to20). The solid line is for the NO-GPSforecasts, and the dashed line is for theGPS forecasts.

1 2 3 4 5Time (day)

0.6

0.8

1

1.2

1.4

1.6

RM

S e

rror

(g/

kg)

1 2 3 4 56

8

10

12

14

16

RM

S e

rror

(m

)

(a)

(b)

Fig. 12. Same as Fig. 9a–b except for theNorthern Hemisphere tropics (0�–30�N).


dotted line). As seen clearly, forecasts ini-tialized with either NO-GPS or GPS analysesare not as skillful as the NCEP operationalforecast due to the omission of AMSU radiancedata in these two experiments. However, theaddition of CHAMP GPS RO observations re-duces the rms forecast error in the Southern

Hemisphere even when the NCEP operationalanalyses are used for verification.

5. Summary and conclusions

The German/US CHAMP mission, carrying anew generation GPS receiver, has providedabout 200 RO observations daily since Febru-ary 11, 2001. Almost fifty percent of CHAMPoccultations descend to a height lower than850 hPa, which alleviates one of the limitationsof the GPS/MET experiments. The CHAMP ob-servations allow the added value of RO obser-vations to global analyses and forecasts to beassessed using a recently improved observationoperator and the 3D-Var system for assimilat-ing GPS RO bending angles.

The impact of RO observations on globalanalyses and forecasts are assessed in thepresence of other observations (excluding theNOAA-16 AMSU radiances) for the first time.The NCEP SSI analysis system is used for dataassimilation, and the NCEP global spectralmodel is used for forecasts. Both data assimila-tion and forecast experiments are conducted atT170L42 resolution. CHAMP RO bending an-gles from July 16th to July 31st, 2002 are in-cluded in the NCEP SSI analysis system. A to-tal of 3030 occultations that were processed byUCAR CDAAC during this period are includedin a two-week data assimilation cycle. Of the3030 total CHAMP profiles, 229 occurred with-in a G3-h time window and a 200-km distanceof at least one radiosonde or dropsonde pro-file. These collocated in-situ observations areexcluded from both data assimilation experi-ments for evaluation purposes. Analyses pro-duced with CHAMP bending angle observa-tions are compared with analyses producedwithout CHAMP observations. The followingare major findings:

1. The analyzed temperatures in the SouthernHemisphere become higher when CHAMPdata are included. Since the analyses with-out CHAMP observations have a cold biaserror in the Southern Hemisphere, inclusionof CHAMP observations reduces this coldbias significantly.

2. When CHAMP observations are included,the analyzed water vapor decreases in thelower troposphere (below 700 hPa) and in-creases in the middle troposphere (400–

16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32July 2002

0

20

40

60

80

100

120R

MS

err

or (

m)

1 day

2 day

3 day

4 day

5 day

Fig. 13. Daily variation of the RMS errorof the 500-hPa height for forecast daysone, two, three, four and five from July16th to July 31st, 2002 in the SouthernHemisphere. The solid line is for theNO-GPS forecasts, and the dashed lineis for the GPS forecasts.

0 1 2 3 4 5Time (day)

0

10

20

30

40

50

60

70

80

90

100

RM

S e

rror

(m

)

NO-GPSGPSNCEP

Fig. 14. The RMS errors of 500 hPa geo-potential height forecasts in the South-ern Hemisphere from one to five daysverified against the NCEP operationalanalyses. The solid line is for the NO-GPS forecasts, the dashed line is for theGPS forecasts, and the dash-dotted lineis for the NCEP operational forecasts.


700 hPa), particularly in the mid-latitudesof the Southern Hemisphere, the tropics andthe subtropics. Adding CHAMP observationsin both the Southern Hemisphere and NorthAmerica increases the mean amount ofwater vapor in the analyses compared toindependent soundings above 800 hPa anddries the analyses compared to the sound-ings below this level.

3. Inclusion of CHAMP observations does notproduce significant changes in the averagesurface pressure analysis in the NorthernHemisphere. However, a slight increase ofthe average surface pressure (maximum1 hPa) is observed in the Southern Hemi-sphere.

4. One- to five-day forecasts initialized withand without CHAMP observations demon-strate that the forecast skill is improvedslightly with the inclusion of CHAMP datain the Southern Hemisphere and in theNorthern Hemisphere tropics.

It appears that the number of CHAMP ob-servations is still too small (about 25 occulta-tions within a 6-h time window in the NorthernHemisphere) to make a significant difference inthe Northern Hemisphere analyses. Problemsassociated with humidity analysis in the lowertroposphere for GPS data assimilation, whichmay be related to the negative N-bias in someof the CHAMP data, need to be studied further.The COSMIC constellation of six satellites, tobe launched in late 2005, will provide up to3000 soundings per day globally (Rocken et al.2000). The COSMIC mission will enable thetesting of the impact of a large number of ROsoundings on operational NWP in real time.

Acknowledgments

This research is supported by the NationalScience Foundation under the project ATM-0101036 and the Integrated Program Office ofNOAA’s National Polar-orbiting OperationalEnvironmental Satellite System under SMC/CIPN Project Order No. Q000C1737600086.The authors would like to thank Dr. Doug Huntfor providing CHAMP data and technical con-sultation and Dr. Sergey Sokolovskiy for pro-viding us the code of vector rotation. The au-thors would also like to thank Dr. Steve Lord atNCEP for his support to use the NCEP SSI

system in this work. Finally, we thank Dr.Y.-H. Kuo for his review of the manuscript andhelpful comments.

Appendix A

Computation of the normal vectorof the occultation plane using CHAMP

level-2 data

The normal vector of an occultation plane is arequired input quantity of the ray-tracing op-erator. It was directly provided in the GPS/MET data sets. Instead of the normal vector,the azimuth angle at the perigee point of anoccultation is provided by the CHAMP datasets. In the following, we first explain the azi-muth angle provided by CHAMP and then de-rive the normal vector of an occultation planefrom its azimuth angle and perigee point posi-tion, which are both provided in the CHAMPlevel-2 data set (see Table 1).

The azimuth angle in CHAMP data is de-scribed as ‘‘the angle from the occultation planein the direction of the GPS satellite with thenorth direction’’. Figure 15 illustrates schemat-ically several important geometrical featuresassociated with a ray path, which goes from aGPS satellite to a LEO satellite, including theazimuth angle. Point Pt is the perigee point.The center of the local curvature is denoted asO. The vector from O to Pt is called the perigeepoint position vector (denoted by rt). A spherewith its origin at O and radius jrj can be drawn.The perigee point Pt is on the surface of thissphere. Two planes are considered. The firstone is the occultation plane, which is defined bythe center of the local curvature and the LEOand the GPS satellite positions. The second oneis a meridian plane, which is defined by thecenter of the local curvature, the perigee pointand the north pole. The occultation plane, themeridian plane and the spherical surface inter-sect at Pt. We introduce two unit vectors start-ing at Pt: u t

occ and u tmer. The vector u t

occ is a unitvector opposite to the tangent direction of theray at the perigee point, i.e., it points from theperigee point in the direction of the GPS satel-lite and is tangent to the sphere at the tangentpoint. The vector u t

mer points from the perigeepoint in the direction of the north pole, and isalso tangent to the sphere at the perigee point.The vector u t

occ lies in the occultation plane and


the vector u tmer lies in the meridian plane. We

notice that the direction of u tmer is the local di-

rection of north at the perigee point. Thus, ac-cording to its definition, the azimuth angle Az

provided by CHAMP data is the angle betweenu t

occ and u tmer (see Fig. 15). If the vector normal

to the occultation plane ðunoccÞ and the vector

normal to the meridian plane ðunmerÞ are defined

at the right hand side of the vectors u tocc and

u tmer, respectively, the angle between these two

normal vectors is equal to Az based on simplegeometrical relationships shown in Fig. 16.

Having introduced the relationships amongu t

occ;utmer;u

nocc;u

nmer, and Az, it is easy to show

how the normal vector unocc can be calculated

from the known azimuth angle and the positionof the perigee point on the sphere. Since theperigee point position vector rt lies in both themeridian plane and the occultation plane, itis perpendicular to both un

occ and unmer. A local

Cartesian coordinate system ðx 0; y 0; z 0Þ can bedefined as shown in Fig. 17. The center of thecoordinates is at the perigee point Pt. Thevertical axis ðz 0Þ is defined by rt, the x 0-axis isdefined by un

occ, and the y 0-axis is 90� anti-clockwise of the x 0-axis in the plane defined byun

occ and unmer. Since un

occ can be obtained by ro-tating un

mer around the z 0-axis clockwise by anangle of Az;u

nocc can be expressed mathemati-

cally as

unocc ¼ R � un

mer; ð3Þ

where R is a 3-dimensional rotation matrixaround the axis z 0 by an angle Az in the clock-wise direction.

ttocc

merAZ

mern

nocc

tt

Fig. 15. Schematic illustration of the azi-muth angle and other vectors used inthe computation of the unit vector nor-mal to the occultation plane. A raypasses through the atmosphere from aGPS satellite to a LEO satellite. Pt isthe perigee point of the ray. O is thecenter of the local curvature. With O asthe origin, a sphere is defined as pass-ing through the tangent point Pt. Twomajor planes, the meridian plane andthe occultation plane, intersect thesphere at the perigee point. The tan-gent vectors at Pt in these two planesare defined as u t

mer and u tocc. u t

mer is theunit tangent vector in the meridianplane, pointing from Pt in the directionof the local north. u t

occ is the unit tan-gent vector in the occultation plane,pointing from Pt in the direction of theGPS satellite. un

mer and unocc are the unit

vectors normal to the meridian planeand the occultation plane, respectively.The angle between u t

mer and u tocc is the

azimuth angle ðAzÞ of the occultationplane defined in the CHAMP data. Itcan be shown that the angle betweenun

mer and unocc is equivalent to the angle

between u tmer and u t

occ.

ut

meru

t

occ

un

mer

un

occ

Az

Az

Fig. 16. A local Cartesian coordinate.The coordinate center is at Pt. The ver-tical axis z 0 is in the direction of rt andthe x 0 � y 0 plane is defined by un

mer andun

occ with the x 0 axis in the direction ofun

occ. The angle between unmer and un

occ

(the x 0 axis) is equal to the azimuth an-gle ðAzÞ of the ray. Therefore, un

occ canbe obtained by rotating un

mer around rt

by an angle Az in a clockwise manner inthe x 0 � y 0 plane.


In a right-handed global Cartesian coordi-nate system defined at the origin O, the threecomponents of the unit perigee point positionvector rt/jrtj are functions of the geodetic lati-tude ðfÞ and longitude ðlÞ of the perigee point:

xt ¼ cos f cos l; ð4Þ

yt ¼ cos f sin l; ð5Þ

zt ¼ sin f: ð6Þ

Similarly, the three components of the unitvector normal to the meridian plane, un

mer is

unmer ¼

264�sin l

cos l

0

375: ð7Þ

The 3-dimensional rotation matrix R is (Glass-ner 1990)

R ¼tx2

t þ c txtyt � szt txtzt þ syt

txtyt þ szt ty2t þ c tytzt � sxt

txtzt � syt tytzt þ sxt tz2t þ c

264

375; ð8Þ

where

c ¼ cos Az; s ¼ �sin Az; t ¼ 1 � cos Az: ð9Þ

Therefore, all values at the right hand side ofthe equation (3) can be derived from givenCHAMP data using (4)–(9).

References

Ao, C.O., T.K. Meehan, G.A. Hajj, and A.J. Man-nucci, 2003: Lower troposphere refractivitybias in GPS occultation retrievals. J. Geophys.Res., 108, 4577.

Engeln, A.v., G. Nedoluha, G. Kirchengast, and S.Buhler, 2003: One-dimensional variational(1-D Var) retrieval of temperature, water va-por, and a reference pressure from radio occul-tation measurements: A sensitivity analysis. J.Geophys. Res., 108, 4337–4349.

Eyre, J.R., 1994: Assimilation of radio occultationmeasurements into a numerical weather pre-diction system. ECMWF, ECMWF Tech.Memo., 199, 34pp.

Fjeldbo, G., G.A. Kliore, and V.R. Eshleman, 1971:The neutral atmosphere of Venus as studiedwith the Mariner V radio occultation experi-ments. Astron. J., 76, 123–140.

Glassner, A.S., 1990: Graphics Gems, AcademicPress, 466pp.

Hajj, G.A., C.O. Ao, B.A. Iijima, D. Kuang, E.R. Kur-sinski, A.J. Mannucci, T.K. Meehan, L.J. Ro-mans, M. de la Torre Juarez, and T.P. Yunck,2003: CHAMP and SAC-C atmospheric resultsand intercomparisons. J. Geophys. Res. Atmos-pheres, (submitted).

Healy, S.B. and J.R. Eyre, 2000: Retrieving temper-ature, water vapor and surface pressure in-formation from refractivity-index profiles de-rived by radio occultation: A simulation study.Quart. J. Roy. Meteor. Soc., 126, 1661–1683.

Kuo, Y.-H., S. Sokolovskiy, R. Anthes, and F. Van-denberghe, 2000: Assimilation of GPS radiooccultation data for numerical weather predic-tion. TAO, 11, 157–186.

———, T.-K. Wee, S. Sokolovskiy, C. Rocken, W.Schreiner, D. Hunt, and R.A. Anthes, 2004: In-version and error estimation of GPS radio oc-cultation data. J. Meteor. Soc. Japan, (sub-mitted).

———, X. Zou, and W. Huang, 1997: The impactof GPS data on the prediction of an extra-tropical cyclone: An observing system simula-tion experiment. J. Dyn. Atmos. Ocean., 27,439–470.

Kursinski, E.R. and G.A. Hajj, 2001: A comparison ofwater vapor derived from GPS occultations andglobal weather analyses. J. Geophys. Res., 106,1113–1138.

Kursinski, E.R., S.B. Healy, and L.R. Romans, 2000:Initial results of combing GPS occultations

z'

rt

y'

un

mer

un

occ

x'

Az

Pt

Fig. 17. Schematic illustration of a localCartesian coordinate with its origin de-fined at the tangent point ðPtÞ. The ver-tical z 0-axis is in the direction of rt, thex 0-axis is defined by un

occ, and the y 0-axisis in the plane defined by un

occ and unmer.

Since the angle between unocc and un

mer isequal to the azimuth angle ðAzÞ, un

occ

(the unknown vector) can be obtainedby rotating un

mer (the known vector)clockwise by an angle Az (known infor-mation) in the x 0 � y 0 plane.


with ECMWF global analyses within a 1D-Varframework. Earth, Planets and Space, 52, 885–892.

Liu, H. and X. Zou, 2003: Improvements to a forwardGPS raytracing model and their impacts onassimilation of bending angle. J. Geophys. Res.,108, doi: 10.1029/2002 JD 003160, 2003.

———, ———, R.A. Anthes, J.C. Chang, J.-H. Tseng,and B. Wang, 2001: The impact of 837 GPS/MET bending angle profiles on assimilationand forecasts for the period June 20–30, 1995.J. Geophys. Res., 106, 31771–31786.

Melbourne, W.G., E.S. Davis, C.B. Duncan, G.A.Hardy, E.R. Kursinski, T.K. Meehan, L.E.Young, and T.P. Yunck, 1994: The applicationof spaceborne GPS to atmospheric limb sound-ing and global change monitoring. NASA andJPL, JPL Publication 94-18, 147pp.

Palmer, P.I., J.J. Barnett, J.R. Eyre, and S.B. Healy,2000: A non-linear optimal estimation inversemethod for radio occultation measurements oftemperature, humidity and surface pressure.J. Geophys. Res., 105, 17513–17526.

——— and ———, 2001: Application of an optimalestimation inverse method to GPS/MET bend-ing angle observations. J. Geophys. Res., 106,17147–17160.

Parrish, D.F. and J. Derber, 1992: The National Me-teorological Center’s spectral and statistical-interpolation analysis system. Mon. Wea. Rev.,120, 1747–1763.

Poli, P., J. Joiner, and E.R. Kursinski, 2002: 1DVARanalysis of temperature and humidity usingGPS radio occultation refractivity data. J. Geo-phys. Res., 107, 4448–4467.

Rocken, C., R. Anthes, M. Exner, D. Hunt, S. Soko-lovskiy, R. Ware, M. Gorbunov, W. Schreiner,D. Feng, B. Herman, Y.-H. Kuo, and X. Zou,1997: Analysis and validation of GPS/METdata in the neutral atmosphere. J. Geophys.Res., 102, 29849–29866.

———, Y.-H. Kuo, W.S. Schreiner, D. Hunt, S.Sokolovskiy, and C. McCormick, 2000: COS-MIC system description. TAO, 11, 21–52.

Shao H. and X. Zou, 2003: On the observational

weighting and its impact on GPS/MET bend-ing angle assimilation. J. Geophys. Res., 107,1–28.

Tsuda, T., M. Mishida, C. Rocken, and R. Ware,2000: A global morphology of gravity wavesactivity in the stratosphere revealed by theGPS occultation data (GPS/MET). J. Geophys.Res., 105, 7257–7273.

Ware, R., M. Exner, D. Feng, M. Gorbunov, K.Hardy, B. Herman, Y.-H. Kuo, T. Meehan, W.Melbourne, C. Rocken, W. Schreiner, S. Soko-lovskiy, F. Solheim, X. Zou, R. Anthes, S. Busi-nger, and K. Trenberth, 1996: GPS sounding ofthe atmosphere from low earth orbit: Prelimi-nary results. Bull. Am. Meteor. Soc., 77, 19–40.

Wickert, J., C. Reigber, G. Beyerle, R. Konig, C.Marquardt, T. Schmidt, L. Grunwaldt, R.Galas, T.K. Meehan, W.G. Melbourne, and K.Hocke, 2001: Atmosphere sounding by GPSradio occultation: First results from CHAMP.Geophys. Res. Lett., 28, 3263–3266.

Yunck, T.P., C.H. Liu, and R. Ware, 2000: A historyof GPS sounding. Terrestrial, Atmospheric andOceanic Science, 11, 1–20.

Zou, X., Y.-H. Kuo, and Y.-R. Guo, 1995: Assimilationof atmospheric radio refractivity using a non-hydrostatic adjoint model. Mon. Wea. Rev.,123, 2229–2249.

———, F. Vandenberghe, B. Wang, M.E. Gorbunov,Y.-H. Kuo, S. Sokolovskiy, J.C. Chang, J.G.Sela, and R. Anthes, 1999: A raytracing opera-tor and its adjoint for the use of GPS/MET re-fraction angle measurements. J. Geophys. Res.,104, 22301–22318.

———, B. Wang, H. Liu, R.A. Anthes, T. Matsu-mura, and Y.-J. Zhu, 2000: Use of GPS/METrefraction angles in 3D variational analysis.Quart. J. Roy. Meteor. Soc., 126, 3013–3040.

———, H. Liu, and R.A. Anthes, 2002: A statisticalestimate of errors in the calculation of radiooccultation bending angles caused by a 2D ap-proximation of raytracing and the assumptionof spherical symmetry of the atmosphere. J.Atmos. Ocean. Tech., 19, 51–64.


impact of champ radio occultation observations on global ...250 k (hajj et al. 2003). the new champ...

Documents